The DI module/library is a port of [di.js](https://github.com/angular/di.js) (+ the best parts of [di.dart](https://github.com/angular/di.dart)) to ES6+A.
## Core Abstractions
The library is built on top of the following core abstractions: `Injector`, `Binding`, and `Dependency`.
* An injector resolves dependencies and creates objects.
* A binding maps a token to a factory function and a list of dependencies. So a binding defines how to create an object. A binding can be synchronous or asynchronous.
* A dependency points to a token and contains extra information on how the object corresponding to that token should be injected.
```
[Injector]
|
|
|*
[Binding]
|----------|-----------------|
| | |*
[Token] [FactoryFn] [Dependency]
|---------|
| |
[Token] [Flags]
```
#### Key and Token
Any object can be a token. For performance reasons, however, DI does not deal with tokens directly, and, instead, wraps every token into a Key. See the section on "Key" to learn more.
In this example we create two bindings: one for Car and one for Engine. `@Inject(Engine)` declares that Car depends on Engine.
## Injector
An injector instantiates objects lazily, only when needed, and then caches them.
Compare
```
var car = inj.get(Car); //instantiates both an Engine and a Car
```
with
```
var engine = inj.get(Engine); //instantiates an Engine
var car = inj.get(Car); //instantiates a Car
```
and with
```
var car = inj.get(Car); //instantiates both an Engine and a Car
var engine = inj.get(Engine); //reads the Engine from the cache
```
To avoid bugs make sure the registered objects have side-effect-free constructors. If it is the case, an injector acts like a hash map with all of the registered objects created at once.
* The type annotation is used by DI only when no @Inject annotations are present.
*`InjectPromise` tells DI to inject a promise (see the section on async for more information).
*`InjectLazy` enables deferring the instantiation of a dependency by injecting a factory function.
## Async
Asynchronicity makes code hard to understand and unit test. DI provides two mechanisms to help with it: asynchronous bindings and asynchronous dependencies.
Suppose we have an object that requires some data from the server.
Both the UserList and UserController classes have to deal with asynchronicity. This is not ideal. UserList should only be responsible for dealing with the list of users (e.g., filtering). And UserController should make ui-related decisions based on the list. Neither should be aware of the fact that the list of users comes from the server. In addition, it clutters unit tests with dummy promises that we are forced to provide.
The DI library supports asynchronous bindings, which can be used to clean up UserList and UserController.
bind(UserList).toAsyncFactory(() => fetchUsersUsingHttp().then((u) => new UserList(u))),
UserController
])
var uc:Promise = inj.asyncGet(UserController);
```
Both UserList, UserController are now async-free. As a result, they are easy to reason about and unit test. We pushed the async code to the edge of our system, where the initialization happens. The initialization code tends to be declarative and relatively simple. And it should be tested with integration tests, not unit tests.
Note that asynchronicity have not disappeared. We just pushed out it of services.
DI also supports asynchronous dependencies, so we can make some of our services responsible for dealing with async.
bind(UserList).toAsyncFactory(() => fetchUsersUsingHttp().then((u) => new UserList(u))),
UserController
])
var uc = inj.get(UserController);
```
We can get an instance of UserController synchronously. It is possible because we made UserController responsible for dealing with asynchronicity, so the initialization code does not have to.